首页|期刊导航|井冈山大学学报(自然科学版)|基于神经网络定位算法的系统设计与仿真

基于神经网络定位算法的系统设计与仿真OA

System design and simulation of neural network-based localization algorithm

中文摘要英文摘要

针对水下小区域内目标定位问题,本研究构建了一套软硬协同的高精度水下定位系统.在硬件端,经过仿真验证后,给出了对应的硬件设计,其中水声信标在理论上将平均工作功耗维持在 76微瓦,成本降至10元级;同时在接收端设计了宽频可变增益放大器(VGA)的动态压缩,以获取高保真波形数据.在算法解算层,提出基于 Slim U-Net 轻量化全卷积网络的直达波到达时间(ToA)概率分布提取方法,结合亚采样技术实现纳秒级时延估计;在此基础上,针对多径造成的非视距(NLOS)观测粗差,级联了融合 Soft-L1 鲁棒损失边界与信赖域约束的混合非线性寻优引擎,有效阻断坐标发散.蒙特卡洛三维声场联合仿真表明,该系统在维持微瓦级功耗的同时,将其近场95%置信区间的动态跟踪误差钳制在 0.040 m,全局全域 RMSE 稳定在0.111 m,提升了复杂水声环境下空间抗差免疫力与定位精度.

To address the challenges of target positioning in small underwater areas,this paper proposes a high-precision underwater positioning system featuring deep hardware-software synergy.On the hardware side,corresponding hardware designs have been finalized after simulation verification.The underwater acoustic beacon can theoretically maintain an average operating power consumption of 76 microwatts with the costs reduced to the 10-yuan level.Meanwhile,a wideband variable gain amplifier(VGA)is deployed at the receiving terminal to perform dynamic signal compression to acquire high-fidelity waveform data.Algorithmically,a lightweight Slim U-Net-based method is proposed to extract the probability distribution of the direct wave Time of Arrival(ToA).Combined with sub-sampling,this achieves nanosecond-level delay estimation.To mitigate multipath-induced non-line-of-sight(NLOS)outliers,a hybrid nonlinear optimization engine integrating a Soft-L1 robust loss boundary and trust-region constraints is cascaded,which effectively eliminate positioning coordinate divergence.Monte Carlo 3D acoustic field simulations demonstrate that,while sustaining milliwatt-level power consumption,the system tightly constrains the near-field dynamic tracking error(95%confidence interval)at 0.040 m,and stabilizes the global RMSE at 0.111 m.This significantly enhances the spatial immunity in outliers and positioning accuracy for complex underwater environments.

陈磊;郑红

安徽建筑大学电子信息与工程学院,安徽,合肥 230601安徽建筑大学电子信息与工程学院,安徽,合肥 230601||南京信息工程大学海洋科学学院,江苏,南京 210044

通用工业技术

信号与信息处理水下定位系统到达时间差U-Net神经网络多途抑制

signal and information processingunderwater positioning systemtime difference of arrivalU-Net neural networkmultipath suppression

《井冈山大学学报(自然科学版)》 2026 (5)

81-88,8

国家重点研发计划海洋环境安全保障项目(2017YFC1403500)

10.3969/j.issn.1674-8085.2026.05.009

评论